Cisco platform for distributed AI workloads

Informatec Digital » Resources » Cisco Platform for Distributed AI Workloads

The explosion of artificial intelligence and AI agents , and the rise of artificial neural networks , is turning the way we design networks, data centers, and computing platforms on its head. Many organizations have already been confronted with this reality: pilot projects work in the lab, but when it comes to scaling, the infrastructure falls short, bottlenecks appear everywhere, and latency makes real-time decision-making impossible.

In this context, Cisco has been building a comprehensive ecosystem of platforms for distributed workloads , both traditional and AI-based, ranging from the network edge to the interconnection between large data centers for massive clusters. Three pillars stand out within this approach: the new integrated Cisco Unified Edge platform for distributed AI, hybrid management platforms such as Cisco Intersight and Nexus Dashboard , and the new Cisco 8223 with Silicon One P200 , designed for backbone networks that carry massive AI traffic between data centers.

Why current infrastructure falls short for AI

One of the key points Cisco makes bluntly is that more than half of AI pilot projects are stalled due to infrastructure limitations. It's not a lack of ideas or models, but rather a lack of the actual capacity to reliably, securely, and at scale run these workloads.

The problem stems from the fact that AI workloads have shifted from centralized training to real-time distributed inference . Models, and especially AI agents, no longer reside solely in a large cluster in the data center: they are beginning to move to where data is generated and decisions are made, such as stores, factories, hospitals, stadiums, or bank branches.

This change completely transforms network traffic . Instead of more or less predictable bursts, continuous, high-intensity flows are generated. Cisco details that AI agent queries can generate up to 25 times more network traffic than a traditional chatbot , multiplying the pressure on networks, backbones, and computing platforms.

At the same time, the company notes that 75% of corporate data is already created and processed outside the traditional data center , at the edge. Expecting all that information to travel back and forth to the central data center is unrealistic: latency skyrockets, links become saturated, and bandwidth costs skyrocket.

As Jeetu Patel, Cisco's president and chief product officer , summarizes , the current infrastructure is simply not ready to power AI at scale . According to him, computing needs to go where customer interactions and decisions are made: in the branch office, the store, the manufacturing plant, the stadium, or any edge location.

Cisco Unified Edge: The integrated platform for distributed AI workloads

Cisco has introduced Unified Edge as an integrated computing platform specifically designed to run distributed, agent-based AI workloads as close as possible to the data source. It is, in essence, a converged architecture that combines compute, networking, storage, and security into a single edge-oriented system.

The company positions it as a cross-cutting solution for sectors as diverse as retail, healthcare, manufacturing, and financial services . The goal is for it to be equally useful for a chain of stores, a production plant, or a hospital that needs millisecond responses, without always relying on the cloud or a central data center.

According to Cisco, Unified Edge acts as a common foundation to support both traditional workloads and advanced AI use cases . This includes everything from real-time applications that can run on CPUs to GPU-intensive workloads related to complex inference, generative models, or autonomous agents.

A key feature is that the platform is designed to grow without requiring complete replacements . Instead of discarding hardware every few years, the modular architecture allows for expansion of the CPU, GPU, network, or storage capacity as AI needs grow, protecting the initial investment.

To reinforce its practical approach, Cisco co-designed Unified Edge with customers in retail, manufacturing, banking, and healthcare . Input from these sectors has influenced both the technical architecture and how the platform is deployed, secured, and managed at scale, with real-world scenarios, not just in the lab.

Converged architecture and modular chassis: CPU, GPU, network and storage

At the heart of Unified Edge is a converged architecture that unifies compute, networking, and storage within a single physical platform. It's not about adding separate pieces, but about integrating all the building blocks into a common design optimized for the edge.

The system is based on a modular chassis that supports multiple CPU and GPU configurations , allowing processing power to be tailored to each scenario. A small shop can start with a lighter configuration, while an industrial plant using computer vision or a hospital with medical image analysis can opt for more GPUs from day one.

Furthermore, Unified Edge incorporates high-performance SD-WAN networks , designed to efficiently and securely connect edge sites to the corporate network core or the cloud. This connectivity is essential for moving models, data, and telemetry without creating bottlenecks.

The platform also includes redundant power and cooling systems , designed to ensure service continuity in environments where a power outage or thermal failure is not an option. Cisco's pre-validated designs help ensure consistent and repeatable integration of all these elements (computing, storage, networking, power, and cooling).

This modularity, combined with the convergence of resources, allows organizations to deploy edge computing in a scalable way , without needing to completely redesign their infrastructure every time new AI workloads appear or installed capacity needs to be expanded.

Real-time inference and agent-based workloads

One of Cisco's core messages is that Unified Edge is optimized for real-time AI inference and intelligent agent-based workloads . Instead of simply running models centrally, the platform allows them to be deployed close to where critical decisions are made.

This is crucial, for example, in factories that need to detect anomalies in real time , in stores that want to personalize the customer experience instantly, or in healthcare facilities where every millisecond matters in clinical data analysis. In all these cases, continuously sending data to a distant cloud is not viable due to both latency and cost.

Unified Edge is designed to allow workloads to evolve over time . Today it might be running "classic" applications that are primarily CPU-driven, and tomorrow it could serve as the foundation for complex models that leverage GPUs. This hybrid design helps businesses manage current realities while preparing for a much more AI-intensive future.

The architecture covers the entire path, from the edge to the core of the network , so that agents and models can operate locally, but also coordinate with central resources when more power or data aggregation is needed.

This approach reduces the risk of organizations becoming "locked" into a single type of load, while also fostering the emergence of new, yet-to-be-imagined use cases , supported by the same technological base.

Centralized management with Cisco Intersight and automated operations

Deploying hundreds or thousands of edge nodes without a robust control plane would be unmanageable. That's why Cisco has integrated centralized management into Unified Edge through Cisco Intersight , its hybrid cloud management platform already used by many customers for data center and private cloud environments.

The goal is for IT teams to be able to monitor, configure, automate, and update edge infrastructure without having to physically travel to each site. This includes everything from initial provisioning (with virtually "unattended" deployments) to firmware updates, policy management, and incident resolution.

Intersight is complemented by advanced capabilities such as Intersight Kubernetes Service , which simplifies the lifecycle management of Kubernetes clusters and containerized applications, regardless of their location: in the data center, the public cloud, or at the edge. This aligns perfectly with the trend of packaging AI applications and microservices in containers.

The platform also integrates Intersight Workload Optimizer , a tool that provides real-time visibility into resource usage and helps balance performance and cost. Thanks to this feature, IT teams can automatically adjust CPU, memory, or storage usage for each application, preventing over-provisioning or system overload.

Finally, the integration with AppDynamics allows for correlating infrastructure behavior with the user experience at the application level, anticipating problems that could affect the performance of critical services. All of this contributes to making the platform's operation much more proactive than reactive.

End-to-end observability: Splunk and ThousandEyes

For distributed management to truly work, observability is essential. Cisco has integrated connectors with Splunk and ThousandEyes into Unified Edge , two tools that provide detailed visibility into what's happening across the network and systems.

With Splunk, organizations can aggregate metrics, logs, and events from the platform , analyze them in real time, and apply anomaly detection or advanced correlations. This translates into a much greater ability to detect incidents before they become a serious problem.

ThousandEyes, for its part, offers end-to-end visibility into network and digital service performance , including the portion of the internet that is outside the customer's direct control. This is especially useful when AI workloads rely on external cloud services, APIs, or cross-site connections.

The combination of Intersight, Splunk, and ThousandEyes seeks to "democratize" infrastructure management , allowing NetOps, SecOps, DevOps teams and application managers to share a common view of reality, instead of operating in silos.

In practice, this complete observability facilitates edge scaling, network investment prioritization, and data-driven decision-making about where the real bottlenecks are.

Multi-layered security and Zero Trust model at the edge

As computing moves out of the controlled environment of the data center, the attack surface expands significantly in the face of cybersecurity threats . Devices deployed in stores, factories, or branch offices are more exposed to physical risks and cyberattacks.

Cisco has designed Unified Edge with multi-layered security based on the Zero Trust model , where nothing and no one is considered trustworthy by default. This involves constant authentication and authorization, deep segmentation, and consistent policies for every access point and data flow, complemented by enhanced firewalls.

The platform includes hardware-level tamper-proof features , such as detection of attempts to open or alter the system, as well as detailed telemetry that allows for constant monitoring of equipment status and detection of anomalous behavior.

Security policies are applied consistently across all nodes, reducing the likelihood of divergent configurations or security gaps between sites. Audit logs help maintain regulatory compliance when the infrastructure scales to dozens or hundreds of locations.

This built-in security can be extended to apply Zero Trust principles to every application and every AI model , protecting both data and agent logic and inferences. Since many AI operations at the edge handle sensitive data (health, financial, personal), this approach is especially critical.

Agile IT platforms for multicloud and hyper-distributed environments

Beyond Unified Edge, Cisco has spent years building IT platforms designed for a multi-cloud and hyper-distributed world . During the Cisco Partner Summit Digital, the company emphasized that applications are becoming increasingly distributed, employees are more mobile, and the demands on systems continue to rise.

To address this reality, Cisco is driving solutions that simplify operations, offer consistency across public, private, and edge clouds , and connect teams, tools, and infrastructure. The goal is to provide IT teams with a common framework for managing their entire environment while maintaining a focus on application performance.

These solutions include Cisco Intersight as a hybrid management platform , the Nexus Dashboard itself for operating multi-cloud data center networks , and systems such as Cisco Identity Services Engine (ISE) , which extend the Zero Trust approach to network access anywhere.

Intersight's role, as already mentioned, is to become the "world's simplest" hybrid cloud platform , connecting private data centers with public clouds. Its ability to manage Kubernetes, optimize workloads, and correlate infrastructure with application performance is key in this new scenario.

All of this fits with Cisco's overall strategy of offering platforms, not just isolated products , helping organizations respond to disruptions, accelerate cloud adoption, and transform their IT operating models to support AI and modern applications.

Cisco Nexus Dashboard and Cisco ISE: Multicloud networking and secure access

Within this strategy, Cisco Nexus Dashboard stands out as a unified information and automation interface for multi-cloud data center networks . Its mission is to orchestrate, monitor, and secure networks that extend from on-premises facilities to virtual and edge environments.

Nexus Dashboard consolidates orchestration, analytics, and security services into a single platform , facilitating the adoption of cloud-native applications and accelerating the return on infrastructure investment. This unified view is essential for organizations connecting their AI clusters and distributed workloads across multiple clouds.

Cisco Identity Services Engine (ISE) acts as a central element to simplify secure network access across all domains . It intelligently identifies a wide variety of devices, including IoT endpoints, and enforces consistent policies from the cloud.

With ISE, companies can extend the zero-trust workplace concept to any location and device type , which is critical as computing and AI move to the edge and endpoints multiply.

Together, these pieces reinforce Cisco's vision of a coherent IT platform, where networking, security, and computing work in a coordinated manner to support traditional and AI workloads without creating unnecessary layers of complexity.

Cisco 8223 and Silicon One P200: the backbone for AI workloads

As edge computing gains prominence, large AI clusters continue to grow in size and power consumption. This is where the new Cisco 8223 comes in, a routing system optimized to connect data centers and support the massive traffic generated by training and massive inference workloads.

The Cisco 8223 is, according to the company itself, the industry's only 51,2 Tbps fixed Ethernet router designed for AI traffic between data centers . It is powered by the Cisco Silicon One P200 chip , which acts as the system core and provides high bandwidth, energy efficiency, and programmability.

The reason for this team's existence is clear: AI computing already exceeds the capacity of many individual data centers , forcing a horizontal scale and the distribution of workloads across multiple locations. This trend demands interconnections between data centers hundreds of kilometers apart, with maximum performance and reliability.

If the connection points between data centers are not optimized, bottlenecks, performance issues, and inefficient use of energy and resources result . The 8223 aims to solve precisely this, offering large buffer capacity, high port density, and 800G coherent optics options for distances up to 1.000 km.

Manufacturers like Microsoft, Alibaba, and suppliers like Lumen have publicly highlighted that Silicon One and systems like the 8223 allow them to build more scalable, programmable, and efficient backbone networks to support the new wave of AI and cloud workloads.

Energy efficiency, scalability and intelligence in the 8223

One of the Cisco 8223's strengths is its energy efficiency . It's presented as the most efficient routing system on the market for scalable networks , with density and power consumption comparable to a switch, despite including a very large buffer capacity.

In terms of scalability, the system offers 64 800G ports and can process more than 20.000 billion packets per second . Cisco indicates that it is capable of scaling up to 3 exabits per second, placing it in the league of major global interconnection infrastructures.

The 8223 is designed to absorb massive traffic spikes associated with AI model training , maintaining performance and preventing slowdowns. Thanks to the P200's buffering capabilities, it can handle bursts that would otherwise overwhelm the network.

The P200 silicon is also highly programmable , allowing the 8223 system to intelligently adapt to network conditions in real time and support emerging protocols and standards without requiring hardware changes.

On an operational level, the equipment integrates with Cisco's observability platforms, so customers gain detailed information about AI traffic behavior , facilitating problem detection, capacity planning, and continuous performance improvement.

Security and flexibility in backbone networks for AI

In addition to raw performance, the Cisco 8223 incorporates multiple layers of security that span from hardware to software and the network itself. It includes features such as line-rate encryption with quantum-resistant algorithms , built-in security measures, and continuous monitoring tools.

These capabilities are designed to protect AI data traffic between data centers , which often includes both valuable training sets and sensitive results. The combination of cryptographic protection, advanced telemetry, and integration with observability systems helps to secure these critical paths.

Another key feature is deployment flexibility . Initially, the 8223 is offered with support for open-source SONiC distributions, and Cisco has already indicated that support for IOS XR will be available later. This openness allows customers to adapt the system to their preferred operating models.

The P200 silicon is not limited to just the 8223: it can be deployed on modular and disaggregated platforms , offering a coherent network architecture for service provider environments as well as large enterprises and hyperscalers.

Cisco also plans to integrate the P200 into the Nexus portfolio with NX-OS , aiming to provide a common foundation for high-performance AI, data center, and WAN infrastructures, all supported by the same Silicon One ASIC family.

An ecosystem of partners and customers that drives the adoption of distributed AI

A key differentiator in Cisco's strategy is its broad ecosystem of technology partners, managed services providers, ISVs, and resellers , which act as a lever to bring these platforms to market in a practical way.

In the case of Unified Edge, the company has collaborated closely with customers in retail, manufacturing, financial services and healthcare to adapt the platform to real-world scenarios: from plants with multiple production lines to distributed bank branches or chains of clinics and hospitals.

Partners such as Intel, Verizon, Rockwell Automation, and World Wide Technology have emphasized that the edge has become the new frontier for AI and that integrated solutions like Unified Edge are key to bringing advanced computing and security capabilities directly to where data is generated.

Intel, for example, emphasizes how combining its silicon with Cisco's computing and networking expertise allows the full power of the data center to be extended to the edge , while Verizon highlights the importance of building future-proof edge platforms with simplicity, reliability, and consistency.

Industry analysts point out that the true potential of AI will be unleashed when inference and analysis can naturally move to the edge , right where the data originates, and agree that Cisco's approach—integrating computing, networking, and security into modular solutions—is moving in that direction.

This entire Cisco offering—with Unified Edge for the edge, Intersight and Nexus Dashboard for multicloud management, ISE for secure access, and 8223 with Silicon One P200 for high-capacity backbones—paints a fairly comprehensive picture of what a modern AI workload platform should look like : distributed, modular, observable, automated, secure by design, and flexible enough to support businesses from their current CPU-based applications to the most demanding scenarios of intelligent agents supported by GPUs and advanced models.